Machine Learning for Evaluating the Heterogeneous Effects of Intensive In-Hospital Rehabilitation During the Postacute Phase After Hip Fracture Surgery on Activities of Daily Living
Takaaki Ikeda, Takuaki Tani, Upul Cooray, Yuta Suzuki, Yosuke Kimura, Kazuhiro Miyata, Masayasu Murakami, Ken Osaka, Kiyohide FushimiBackground:
It remains unclear which patient subgroups benefit most from intensive rehabilitation therapy.
Objectives:
This study aimed to examine the heterogeneity in the effects of intensive postacute rehabilitation after hip fracture surgery on activities of daily living and to identify its potential sources.
Research Design:
A retrospective observational study.
Subjects:
Patients aged 50 years or older who underwent hip fracture surgery within 2 days of admission and were transferred to a rehabilitation ward between 7 and 30 days after surgery.
Measures:
Exposure was defined as the average daily rehabilitation time within 30 days posttransfer, dichotomized as ≥120 versus <120 minutes/day. The outcome was the motor domain score of the Functional Independence Measure (FIM) at 60 days posttransfer. Heterogeneity of treatment effects was assessed using conditional average treatment effects (CATEs) estimated through the causal forest approach. The models were adjusted for relevant covariates, including age, sex, and FIM score at the time of transfer to the rehabilitation ward.
Results:
The causal forest approach revealed heterogeneous effects—the estimated CATE for the top 20% high-benefit subgroup was 3.89 (95% CI: 2.32–5.46). The high-benefit subgroup was older and had lower FIM scores for the self-care, mobility, transfers, and excretion control domains at the time of transfer, compared with the low-benefit subgroup (the bottom 20% of the estimated CATEs).
Conclusions:
Heterogeneity exists in the association between intensive in-hospital rehabilitation therapy and functional status among older patients undergoing early hip fracture surgery during the postacute rehabilitation phase.